Introduction
Recent years have seen remarkable advancements in data-driven approaches for observing, modeling, and understanding the Earth system. These methods, which encompass artificial intelligence (AI), machine learning, and empirical dynamical approaches, are transforming our ability to analyze complex atmospheric and oceanic processes across multiple timescales. While physics-based models remain central to Earth system science, data-driven techniques offer new ways to extract meaningful patterns, improve predictability, and develop computationally efficient representations of dynamical processes.
One of the key strengths of data-driven models is their capacity to learn from large datasets, enabling rapid and cost-effective forecasting solutions. In particular, some tasks traditionally performed by numerical models, such as weather and climate predictions, are increasingly complemented—or even challenged—by advanced data-driven methodologies. For example, recent AI-based forecasting systems trained on reanalysis data, such as GraphCast, have demonstrated competitive skill compared to traditional physics-based simulations. These advances highlight the growing interplay between empirical and physical modeling approaches, where hybrid techniques, such as Physics-Informed Neural Networks (PINNs) and explainable AI (XAI), bridge the gap between purely data-driven and physically constrained methods. Moreover, transfer operator-based techniques, including Koopman mode decomposition, offer promising avenues for extracting coherent structures, identifying climate modes, and characterizing the nonlinear dynamics of the climate system.
This special collection in the open-access journal Environmental Data Science (Cambridge University Press) aims to explore the convergence of data-driven methodologies with physical modeling in Earth system science. We welcome contributions that investigate the application of these methods in a wide range of domains, including but not limited to:
- Climate predictability and forecasting
- Spatiotemporal feature extraction
- Climate mode identification and network analysis
- Nonlinear dynamical system analysis and attractor reconstruction
- Development of reduced-order models
- Extreme event detection and attribution
- AI-driven and empirical parameterization of physical models
- Hybrid modeling approaches combining data-driven and physics-based techniques
- Explainable AI (XAI) for improving process-level understanding
Submissions are encouraged to explore both the competitive and complementary aspects of data-driven and physical modeling approaches, particularly in applications related to atmospheric, oceanic, and coupled Earth system processes. Contributions spanning satellite remote sensing, biogeochemistry, cryosphere studies, and societal impacts are also welcome. By fostering discussions at the intersection of data-driven and physics-based methodologies, this special issue aims to advance our ability to model and predict the Earth system with greater accuracy and efficiency.
Timetable
- An EGU 2025 Workshop on Data-driven and physical climate modelling takes place on 2nd May 2025 in Vienna
- Deadline to submit to EDS: 31 October 2025.
We encourage submissions both from authors participating in the workshop and those not taking part. Articles will be peer-reviewed according to the standard EDS process (see below) and will be published as soon as possible after acceptance, in the interest of allowing authors to disseminate their work without unnecessary delay, and added to a special collection page.
How to Submit
Please note the following key details, with more information available in the EDS Instructions for Authors:
Article Types: Authors should make sure that they select the most appropriate article type when they submit their work to EDS. The standard EDS article types are:
- Application papers: Research progress, or tackling a real-world problem, in an environmental field, enabled by data science. For example, AI or data science could be used for understanding of environmental processes, or improving forecasting tools.
- Methods papers: Novel data science methodology inspired by an environmental problem or application. Typically the methodology should be demonstrated in one or more environmental applications.
- Data papers that describe in a structured way, with a short narrative and accompanying metadata, important and re-usable environmental data sets that reside in publicly accessible repositories. These papers promote data transparency and data re-use.
- Survey papers: providing a systematic overview of a method, tool or approach, or a field or subfield that is relevant to environmental data science.
Templates: Authors have the option of using the following EDS templates to help structure their submission:
- LaTeX template files
- Word template
- Overleaf (a collaborative authoring tool based on LaTeX; authors are able to submit directly to through Overleaf; read more about benefits here)
Authors not using our templates are reminded to include:
- An impact Statement: 120 words beneath the abstract describing the significance of the findings in language that can be understood by a wide audience
And at the back of the article:
- Author contributions (using the CRedIT taxonomy as a guide)
- Competing interest statement
- Data availability statement
- Funding statement
See the EDS Instructions for Authors for more details about these statements.
Submission portal: Authors should submit via the EDS ScholarOne site and select the 'Data-driven and Physical Climate Modelling' tag from the dropdown menu when prompted to identify whether the article is for a special collection.
Open Access
Any author can publish on an open access basis in EDS if accepted, irrespective of their funding situation or institutional affiliation. There are no financial barriers to publication. Many articles have publishing costs covered through the Transformative Agreements that Cambridge has set up with universities worldwide. If the corresponding author on an article is affiliated with a Transformative Agreement this effectively covers publishing costs. Authors not affiliated with these agreements who have received a grant that budgets for open access publication are encouraged to pay an article processing charge (APC). However, if an author has no funding and no institutional agreement, the charge will be waived. Please do not let concerns about your financial situation or affiliation put off your submission.
Open Materials
Authors are asked to make code and data that supports the findings openly available in a recognised repository and to link to them in the Data Availability Statement in the article. We recognise this may not be possible in all circumstances. See the EDS Research Transparency policy. Open Data and Open Materials badges will be displayed on published articles that link to replication materials, as a recognition of open practices.
Guest Editors
- Julien Brajard (Nansen Center NERSC, Norway & Sorbonne University, France) - EDS Editorial Board
- Paula Lorenzo Sánchez (University of Bologna, Italy)